The U.S. Department of Energy (DOE) has awarded Dell Technologies a major contract to build NERSC-10—its next flagship supercomputer—at Lawrence Berkeley National Laboratory. Slated to come online in 2026, the new system, named Doudna in honor of CRISPR pioneer and Nobel laureate Jennifer Doudna, will be powered by Dell servers and NVIDIA’s forthcoming Vera Rubin platform, designed to handle large-scale simulation, data, and AI workloads.
Doudna will deliver over 10 times the performance of the current Perlmutter system at the National Energy Research Scientific Computing Center (NERSC). It will support AI training and inference at scale, advanced quantum simulation via CUDA-Q, and next-gen fusion research. The system will feature Dell’s latest liquid-cooled server architecture, NVIDIA’s Blackwell-based GPU-CPU nodes, and Quantum-X800 InfiniBand networking. A reconfigurable heterogeneous compute environment will enable real-time interactive computing and seamless data workflows across DOE facilities through ESnet.
The DOE highlighted the system as a strategic asset in the global AI race, with Secretary of Energy Chris Wright stating, “AI is the Manhattan Project of our time, and Doudna will help ensure America’s scientists have the tools they need to win the global race for AI dominance.” The system is expected to empower over 11,000 users across 800 projects with capabilities that compress “years of discovery into days,” according to NVIDIA CEO Jensen Huang.
- DOE awards Dell Technologies the contract to build NERSC-10, named Doudna
- System to deliver 10x performance over current Perlmutter supercomputer
- Built with Dell ORv3 liquid-cooled servers and NVIDIA’s Vera Rubin platform
- Enables high-performance computing, AI training/inference, and quantum simulation
- Fully integrated with DOE science facilities via ESnet for near real-time workflows
“Doudna is a time machine for science — compressing years of discovery into days,” said Jensen Huang, founder and CEO of NVIDIA.
- NVIDIA’s Vera Rubin platform, named after the astronomer who pioneered dark matter research, is the successor to the Blackwell architecture, set to launch in the second half of 2026. It combines the Rubin GPU and Vera CPU, manufactured using TSMC’s 3nm process, and features 8 stacks of HBM4 memory (up to 288GB) for enhanced AI and HPC workloads. The platform includes NVLink 6 switches with speeds up to 3600 GB/s and CX9 SuperNIC at 1600 Gb/s. The Vera Rubin NVL144 rack delivers 3.6 exaFLOPS of FP4 inference and 1.2 exaFLOPS of FP8 training, offering 3.3x the performance of the GB300 NVL72. A Rubin Ultra variant with 12 HBM4 stacks is planned for 2027, targeting even higher performance with up to 15 exaFLOPS of FP4 compute in the NVL576 configuration.
